The network science approach to discourse analysis
Jun Oshima, Yusuke Niihara, Kensuke Ota, & Ritsuko Oshima,Shizuoka University, Japan
While we have to cope with a large amount of data in the discourse analysis, studies usually show some pieces of discourse with narratives clarifying contexts of learning. This may be an appropriate strategy in describing research findings with limited pages of journals. There are, however, always problems with arbitrary interpretations on discourse by researchers. Even with the same data set, different discourse analysis has different perspectives on what happened in the situation. The purpose of our study is to propose a numerical approach to discourse analysis particularly for providing some objective interpretation of narratives.
The method we apply to discourse analysis is based on the network science. In the network science, structural properties of a variety of complexity systems are explored through statistical analyses and simulations (e.g., Strogatz, 2001; Wattz, & Strogatz, 1998). In this study, we analyzed discourse across four days by a group of five undergraduate students in a course for teacher education. They were selected for the analysis because the instructor evaluated them as the best group in the course. The students were required of discussing their understandings of educational theories in lectures by the instructor. Their discourse during the course was video-recorded and transcribed. Based on the corpus of utterances, we created a network of utterances. Each utterance was independently evaluated by the supportive instructor as “cognitively important” in that the utterance included important propositional knowledge or manifesting its appropriate application, or “not important.”
The betweeness centrality coefficient of every utterance was calculated for evaluating how each utterance played a role of advancing the discourse by mediating other utterances during the discourse. Figure 1 shows time-serial change in means of the betweeness centralities of cognitively important utterances by five learners (A ~ E). Results are summarized as follows. First, every learner was not equally contributing to discourse as expected. Characteristics of each learner’s contribution to the discourse based on the network analysis were found to fit with informal evaluations by the instructor and teaching assistants. Second, surprisingly, learners who constantly manifested high betweeness coefficients did not acquire higher scores in their final essays. The instructor gave the highest scores to learner D and E, but not to A, B, and C. Thus, results of the utterance network analysis with the final essay evaluation suggested us that the network analysis might help us catch the macro level of characteristics of groups and their members in contribution to their discourse in collaborative learning. In addition, the analysis in this study gave us an interesting fact that students who cognitively and actively engaged in collaborative discourse did not necessarily improve their final comprehension on the course contents. For discussing this contradictory result, we need more careful analysis of discourse. Now, we receive this result as a warning for us to search for the correlation-based mechanism of learning process and learning outcome. .
The knowledge building activity was checked in terms of Epistemic Agency (EA hereinafter), through the Content Analysis of the KF notes. A coding scheme was employed to distinguish the contents of the notes in terms of Advanced EA (Exploring Problems and Evaluating contents and strategies) and Basic EA (Proposing and Elaborating Information), with an agreement between two independent judges of 80%.
The results show no changes considering the people presenting a central level of participation. The peripheral participants attending the course with a metacognitive reflection activity start at the beginning with a strong difference between Basic and Advanced EA and at the end of the course they reach a balanced situation between the two kind of EA. The theoretical and practical implications of these results are discussed.
Figure 1. Time-serial change in means of the betweeness centralities of cognitively important utterances by learners.